[YARN-9111] NM crashes because Fair scheduler promotes a container that has not been pulled by AM

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: YARN-1011
• Component/s: fairscheduler, nodemanager
• Assignee: Unassigned
• Reporter: Haibo Chen
• Created: 11/Dec/18 22:38
• Updated: 11/Dec/18 22:38

Description

2018-10-19 22:34:35,052 FATAL org.apache.hadoop.yarn.event.AsyncDispatcher: Error in dispatcher thread
 java.lang.NullPointerException
 at org.apache.hadoop.yarn.server.utils.BuilderUtils.newContainerTokenIdentifier(BuilderUtils.java:323)
 at org.apache.hadoop.yarn.server.nodemanager.containermanager.ContainerManagerImpl.handle(ContainerManagerImpl.java:1649)
 at org.apache.hadoop.yarn.server.nodemanager.containermanager.ContainerManagerImpl.handle(ContainerManagerImpl.java:185)
 at org.apache.hadoop.yarn.event.AsyncDispatcher.dispatch(AsyncDispatcher.java:197)
 at org.apache.hadoop.yarn.event.AsyncDispatcher$1.run(AsyncDispatcher.java:126)
 at java.lang.Thread.run(Thread.java:748)
 2018-10-19 22:34:35,054 INFO org.apache.hadoop.yarn.event.AsyncDispatcher: Exiting, bbye..
 2018-10-19 22:34:35,059 DEBUG org.apache.hadoop.service.AbstractService: Service: NodeManager entered state STOPPED

 

 
When a container is allocated by RM to an application, its container token is not generated until the AM pulls that container from RM.

However, it the scheduler decides to promote that container before it is pulled by the AM, it does not have container token to work with.

The current code does not update/generate the container token as such. When container promotion is sent to NM to process, the NM crashes on NPE.